Re: [PATCH v2 0/3] diff: build parseopts array on demand

From: "Ævar Arnfjörð Bjarmason" <avarab@gmail.com>
To: "René Scharfe" <l.s.r@web.de>
Cc: Git List <git@vger.kernel.org>,
	Junio C Hamano <gitster@pobox.com>, Jeff King <peff@peff.net>
Subject: Re: [PATCH v2 0/3] diff: build parseopts array on demand
Date: Thu, 01 Dec 2022 17:54:31 +0100	[thread overview]
Message-ID: <221201.86r0xjgicc.gmgdl@evledraar.gmail.com> (raw)
In-Reply-To: <ea838ae7-b635-d4d2-d9df-e96b3d8980af@web.de>


On Thu, Dec 01 2022, René Scharfe wrote:

> Calling repo_init_revisions() and release_revisions() in that order
> leaks the memory allocated for the parseopts array in the embedded
> struct diff_options member.  Get rid of that leak by reducing the
> lifetime of that array.
>
> Original patch:
> https://lore.kernel.org/git/4fd82dc6-e0f8-0638-5b10-16bfef39a171@web.de/
>
> Submitted separately from that thread because it's independent enough.
>
> Change since v1:
> - Actually remove the parseopts member.  Its removal got lost during
>   refactoring in v1.  Thank you for spotting that, Junio!
>
>   diff: factor out add_diff_options()
>   diff: let prep_parse_options() return parseopt array
>   diff: remove parseopts member from struct diff_options
>
>  builtin/range-diff.c |  2 +-
>  diff-no-index.c      |  3 +--
>  diff.c               | 26 +++++++++++++++-----------
>  diff.h               |  2 +-
>  4 files changed, 18 insertions(+), 15 deletions(-)
>
> Range-Diff gegen v1:
> 1:  630f95320f = 1:  4dc8b2632b diff: factor out add_diff_options()
> 2:  4b56fa795c = 2:  10903d355e diff: let prep_parse_options() return parseopt array
> 3:  7e54e4370a ! 3:  24bd18ae79 diff: remove parseopts member from struct diff_options
>     @@ diff.c: void diff_free(struct diff_options *options)
>       }
>
>       void diff_flush(struct diff_options *options)
>     +
>     + ## diff.h ##
>     +@@ diff.h: struct diff_options {
>     + 	unsigned color_moved_ws_handling;
>     +
>     + 	struct repository *repo;
>     +-	struct option *parseopts;
>     + 	struct strmap *additional_path_headers;
>     +
>     + 	int no_free;

This looks good to me. Would you mind running the tests with:

	GIT_TEST_PASSING_SANITIZE_LEAK=check GIT_TEST_SANITIZE_LEAK_LOG=true make SANITIZE=leak 

And then marking up the ones that now pass with
TEST_PASSES_SANITIZE_LEAK=true. I think it's all except one of these
(one isn't marked on "master", I forget which one):

	Test Summary Report
	-------------------
	t1022-read-tree-partial-clone.sh                 (Wstat: 256 Tests: 1 Failed: 0)
	  Non-zero exit status: 1
	t2012-checkout-last.sh                           (Wstat: 256 Tests: 22 Failed: 0)
	  Non-zero exit status: 1
	t3210-pack-refs.sh                               (Wstat: 256 Tests: 30 Failed: 0)
	  Non-zero exit status: 1
	t4053-diff-no-index.sh                           (Wstat: 256 Tests: 19 Failed: 0)
	  Non-zero exit status: 1
	t5554-noop-fetch-negotiator.sh                   (Wstat: 256 Tests: 1 Failed: 0)
	  Non-zero exit status: 1
	t5613-info-alternate.sh                          (Wstat: 256 Tests: 13 Failed: 0)
	  Non-zero exit status: 1
	t6021-rev-list-exclude-hidden.sh                 (Wstat: 256 Tests: 42 Failed: 0)
	  Non-zero exit status: 1
	t6415-merge-dir-to-symlink.sh                    (Wstat: 256 Tests: 24 Failed: 0)
	  Non-zero exit status: 1
	t7403-submodule-sync.sh                          (Wstat: 256 Tests: 18 Failed: 0)
	  Non-zero exit status: 1
	t7504-commit-msg-hook.sh                         (Wstat: 256 Tests: 30 Failed: 0)
	  Non-zero exit status: 1
	t9115-git-svn-dcommit-funky-renames.sh           (Wstat: 256 Tests: 12 Failed: 0)
	  Non-zero exit status: 1
	t9146-git-svn-empty-dirs.sh                      (Wstat: 256 Tests: 14 Failed: 0)
	  Non-zero exit status: 1
	t9160-git-svn-preserve-empty-dirs.sh             (Wstat: 256 Tests: 12 Failed: 0)
	  Non-zero exit status: 1

I.e. this makes a lot more tests pass leak-free, yay!

To nitpick a bit: I didn't find that splitting this up into three
patches helped to read it, e.g. 2/3 adds code that promptly goes away in
3/3.

I also wondered why add two API functions for this, instead of just
teaching the "prep options" to concat passed-in options with the user
options? That also avoids a parse_options_dup(). I.e. with that squashed
in the whole thing is:

 builtin/range-diff.c |  2 +-
 diff-no-index.c      |  3 +--
 diff.c               | 19 ++++++++-----------
 diff.h               |  3 ++-
 4 files changed, 12 insertions(+), 15 deletions(-)

diff --git a/builtin/range-diff.c b/builtin/range-diff.c
index e2a74efb42a..8fcd6663b89 100644
--- a/builtin/range-diff.c
+++ b/builtin/range-diff.c
@@ -47,7 +47,7 @@ int cmd_range_diff(int argc, const char **argv, const char *prefix)
 
 	repo_diff_setup(the_repository, &diffopt);
 
-	options = parse_options_concat(range_diff_options, diffopt.parseopts);
+	options = add_diff_parseopts(range_diff_options, &diffopt);
 	argc = parse_options(argc, argv, prefix, options,
 			     builtin_range_diff_usage, PARSE_OPT_KEEP_DASHDASH);
 
diff --git a/diff-no-index.c b/diff-no-index.c
index 18edbdf4b59..efac1d38b38 100644
--- a/diff-no-index.c
+++ b/diff-no-index.c
@@ -255,8 +255,7 @@ int diff_no_index(struct rev_info *revs,
 	};
 	struct option *options;
 
-	options = parse_options_concat(no_index_options,
-				       revs->diffopt.parseopts);
+	options = add_diff_parseopts(no_index_options, &revs->diffopt);
 	argc = parse_options(argc, argv, revs->prefix, options,
 			     diff_no_index_usage, 0);
 	if (argc != 2) {
diff --git a/diff.c b/diff.c
index 1054a4b7329..e186fc91802 100644
--- a/diff.c
+++ b/diff.c
@@ -4615,8 +4615,6 @@ static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
 	builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
 }
 
-static void prep_parse_options(struct diff_options *options);
-
 void repo_diff_setup(struct repository *r, struct diff_options *options)
 {
 	memcpy(options, &default_diff_options, sizeof(*options));
@@ -4662,8 +4660,6 @@ void repo_diff_setup(struct repository *r, struct diff_options *options)
 
 	options->color_moved = diff_color_moved_default;
 	options->color_moved_ws_handling = diff_color_moved_ws_default;
-
-	prep_parse_options(options);
 }
 
 static const char diff_status_letters[] = {
@@ -4821,8 +4817,6 @@ void diff_setup_done(struct diff_options *options)
 			options->filter = ~filter_bit[DIFF_STATUS_FILTER_AON];
 		options->filter &= ~options->filter_not;
 	}
-
-	FREE_AND_NULL(options->parseopts);
 }
 
 int parse_long_opt(const char *opt, const char **argv,
@@ -5419,7 +5413,8 @@ static int diff_opt_rotate_to(const struct option *opt, const char *arg, int uns
 	return 0;
 }
 
-static void prep_parse_options(struct diff_options *options)
+struct option *add_diff_parseopts(struct option *useropts,
+				  struct diff_options *options)
 {
 	struct option parseopts[] = {
 		OPT_GROUP(N_("Diff output format options")),
@@ -5689,22 +5684,25 @@ static void prep_parse_options(struct diff_options *options)
 		OPT_END()
 	};
 
-	ALLOC_ARRAY(options->parseopts, ARRAY_SIZE(parseopts));
-	memcpy(options->parseopts, parseopts, sizeof(parseopts));
+	return parse_options_concat(useropts, parseopts);
 }
 
 int diff_opt_parse(struct diff_options *options,
 		   const char **av, int ac, const char *prefix)
 {
+	struct option no_options[] = { OPT_END() };
+	struct option *diff_parseopts = add_diff_parseopts(no_options, options);
+
 	if (!prefix)
 		prefix = "";
 
-	ac = parse_options(ac, av, prefix, options->parseopts, NULL,
+	ac = parse_options(ac, av, prefix, diff_parseopts, NULL,
 			   PARSE_OPT_KEEP_DASHDASH |
 			   PARSE_OPT_KEEP_UNKNOWN_OPT |
 			   PARSE_OPT_NO_INTERNAL_HELP |
 			   PARSE_OPT_ONE_SHOT |
 			   PARSE_OPT_STOP_AT_NON_OPTION);
+	free(diff_parseopts);
 
 	return ac;
 }
@@ -6513,7 +6511,6 @@ void diff_free(struct diff_options *options)
 	diff_free_file(options);
 	diff_free_ignore_regex(options);
 	clear_pathspec(&options->pathspec);
-	FREE_AND_NULL(options->parseopts);
 }
 
 void diff_flush(struct diff_options *options)
diff --git a/diff.h b/diff.h
index fd33caeb25d..56704d3de22 100644
--- a/diff.h
+++ b/diff.h
@@ -394,7 +394,6 @@ struct diff_options {
 	unsigned color_moved_ws_handling;
 
 	struct repository *repo;
-	struct option *parseopts;
 	struct strmap *additional_path_headers;
 
 	int no_free;
@@ -539,6 +538,8 @@ int git_diff_ui_config(const char *var, const char *value, void *cb);
 #define diff_setup(diffopts) repo_diff_setup(the_repository, diffopts)
 #endif
 void repo_diff_setup(struct repository *, struct diff_options *);
+struct option *add_diff_parseopts(struct option *useropts,
+				  struct diff_options *options);
 int diff_opt_parse(struct diff_options *, const char **, int, const char *);
 void diff_setup_done(struct diff_options *);
 int git_config_rename(const char *var, const char *value);
